2. Oc1 _Demand Forecast Electricity

The Malaysian Grid Code Awareness Programme Funded by Akaun Amanah Industri Bekalan Elektrik (AAIBE)

Operating Code 1:

Demand Forecast

By :

NOOR HAFIFI BIN JALAL


Presentation Outline

Introduction

Short Term Forecasting Process in TNB– ISO Processes

– Demand Profile Analysis

Methodology and Tools– Regression using MetrixND

Forecast and Performance


Introduction

• OC 1 is concerned with demand forecasting for operational purposes.

• OC 1 outlines the obligations of GSO and users on the preparation of

Demand Forecast of Active Power, Active Energy and Reactive Power

on the Transmission System.

• Set out time scales within the Operational Planning and Operational

Control periods.


Introduction (cont.)

Operational Demand Forecast covers:• Day-ahead demand forecast – half-hourly• Week-ahead demand forecast – half-hourly• Month to Four Months-ahead demand forecast-hourly• Special day/week/season demand forecast• 5 years-ahead demand forecast

– hourly (based on the long-term demand forecast prepared by the Grid Owner while formulatingtheSystemDevelopmentPlan)


Objectives

• To enable matching of Generation and Demand in operation

• To ensure the provision of data to the GSO for operation

purposes

• To provide for the factors to be taken into account by the GSO

when Demand forecasting is conducted in operation


Scope

Applies to:

• GSO

• AllGeneratorswithCDGUs

• AllGeneratorswithGeneratingUnitsnotsubjecttoDispatchbytheGSO,withtotalon sitegenerationcapacityequaltoorabove30MWandwheretheGSOconsidersitnecessary

• Directly Connected Customers where the GSO considers it necessary

• NetworkOperators

• Distributors

• DirectlyConnectedCustomerswhohaveagreedto participateinDemandcontrol

• InterconnectedParties.


Short Term Forecasting

Process in TNB


Day-ahead Demand Forecast

Consists of half-hourly demand forecast (00:30 to 24:00) and total energy forecast for the next day

• National grid demand forecast

• Regional demand forecasts (Northern, Central, Southern & Eastern regions)

Used in power system operation planning and real time operation:• Scheduling, unit commitment and dispatching of generating units

• Outage management of generating units

• Outage management of transmission grid network

• Power system security assessments

• Fuel requirements and nominations, etc.

The day-ahead forecast is carried out daily by 10 a.m. for the next day’s demand.

This forecast will be reviewed on the day itself, at 8 a.m. and 4 p.m. The forecast for the day can also be updated on half hourly basis at NLDC using ALSTOM forecasting tool.


Day-ahead Demand Forecast ISO Process


Day-aheadDemand Forecast

Daily Unit Commitment Schedule

Daily System Security Assessment

Dispatch(Real-time)

1000 hrs

1200 hrs

1530 hrs


Factors

MAIN INPUT

DEMAND DATA WEATHER DATA


Latest demand From NLDC

Sample of data retrieved from NLDC (28 May 2013)

• Data is available every 10 minutes

• Data taken from generators

• For our forecasting purposes, we will take half-hourly data


Steel mill demand From NLDC

Sample of data retrieved from NLDC (29 May 2013)

• Data is available every 10 minutes

• Data taken from 8 major steel mills which are connected directly to the grid via 132kV and 275kV feeder


List of Steel Mills

1

4

Voltage Company

275kV and 132kV

Megasteel

132kV Orna Steel

132kV Amsteel

132kV Malayawata

132kV Antara Steel

132kV Perwaja Steel

132kV Malaya Steel Work


Weather Data From MMD

Sample of data retrieved from MMD (29 May 2013)

• Send by email very morning before 8 a.m.

• Hourly weather data in Microsoft Word format

• Stored in excel, accessible by forecasting tool MetrixND

Ramalan Suhu dan Jumlah Hujan setiap Jam bagi Subang (Penghantaran melalui emel [email protected] pada 0700 LT)

Tarikh Masa (LT) Suhu(oC) Jumlah hujan(mm)

29 May 2013 0800 25 0

0900 26 0

1000 28 0

1100 30 0

1200 31 0

1300 32 0

1400 33 0

1500 34 0

1600 32 0.2

1700 30 1.0

1800 28 0.4

1900 28 0.6

2000 27 0.8

2100 27 0.1

2200 27 0.1

2300 27 0

0000 27 0

0100 26 0

0200 26 0

0300 26 0

0400 26 0

0500 25 0

0600 25 0

0700 25 0

30 MEI 2013 0800 26 0

0900 28 0

1000 30 0

1100 31 0

1200 32 0

1300 33 0

1400 34 0

1500 33 0

1600 33 0

1700 32 0

1800 31 0

1900 29 0

2000 28 0

2100 28 0

2200 27 0

2300 27 0


Demand Profile Analysis

– Daily demand Profile:• Typical daily demand profile

– Comparison between days of the week demand profiles

• Hot day vs. Wet day demand profile– Weather impact on demand profile

• National holidays impact on electricity demand– Various national holidays impact on demand demand

• State holidays impact on national electricity demand– Impact from state holidays of each region on the demand

• Festive week demand profile– Comparison between two major festivals: Hari Raya and

Chinese New Year

– Steel Mill demand Profile


Daily Demand Profiles

Maximum demand : 15,652 MW 13/5/2013Maximum energy : 345.254 GWh 25/6/13


Hot and Wet Day Profiles

The variation in demand pattern is highly driven by the weather


Public Holidays

Depending on how much they are celebrated


State Holidays

Depending on how much the state consumes electricity

Selangor is the largest electricity consumer in Peninsular Malaysia


Festive Week Profile

• Festive week daily demand Profile:

Hari Raya Puasa 2012


Festive Week Profile

• Festive week Daily demand Profile:

Chinese New Year 2013


Steel Mill demand Profile

• Typical weekday profile (13 May 2013)


Steel Mill demand Profile

• Typical Sunday profile (12 May 2013)


Weekly MD For 2010 - 2013

Weekly Maximum Demand (MD) trend for the past 4 years


Historical & Future MD

Generation Growth Peak Demand Growth MW

GWh % MW % Increase

2003 73,795 7.1% 11,329 5.1% 546

2004 79,022 7.1% 12,023 6.1% 694

2005 83,303 5.4% 12,493 3.9% 470

2006 86,472 3.8% 12,990 4.0% 497

2007 90,283 4.4% 13,620 4.8% 630

2008 94,370 4.5% 14,007 2.8% 387

2009 92,623 -1.9% 14,245 1.7% 238

2010 100,991 9.0% 15,072 5.8% 827

2011 103,354 2.3% 15,476 2.7% 404

2012 106,884 3.4% 15,826 2.3% 350

2013 111,129 4.0% 16,703 5.5% 877

2014 115,115 3.6% 17,275 3.4% 572

2015 118,886 3.3% 17,813 3.1% 538

2016 123,039 3.5% 18,407 3.3% 594

2017 127,258 3.4% 19,008 3.3% 601

2018 131,551 3.4% 19,619 3.2% 611

2019 135,919 3.3% 20,238 3.2% 620

2020 140,359 3.3% 20,867 3.1% 628

Year

Based on the Electricity Demand Forecast May 2013


Week-ahead Demand Forecast

Consists of hourly demand forecast (24 hrs x 7 days)

Used in power system operation planning:• Scheduling and unit commitment of generating units• Outage management of generating units• Outage management of transmission grid network• Power system security assessments• Fuel requirements and nominations, etc.

Conducted weekly before 3 p.m. on Wednesday To be presented in the Weekly Coordination Meeting on

Thursday at 3 p.m.


Week-ahead Demand Forecast


Month to Four Months-ahead Demand Forecast

Consists of hourly demand forecast (24 hrs x 30 days x 4 months)

Used in power system operation planning:• Scheduling and unit commitment of generating units• Outage management of generating units• Outage management of transmission grid network• Power system security assessments• Fuel requirements and nominations, etc.

Conducted monthly before end of every month


5 years-ahead Demand Forecast

Consists of hourly demand forecast (24 hrs x 365 days x 5 yrs)

Used to prepare:

• 5-year Generation Operation Plan for capacity

adequacy planning

• 2-year System Operation Plan for system

operation and security planning

• Fuel requirement

• Resource planning


Methodology


Introduction

• Quantitative methods have been shown to be helpful in making better predictions about the future course of events.

• However, there is a rich history of forecasting based on subjective or judgmental methods, most of which remain useful today.

• Combine original quantitative forecast with the corresponding judgmentally adjusted forecasts.

• Implies 3 steps: Make quantitative forecast, allow for judgmental adjustment, then combine both forecasts.


Linear Regression

• Specification of the Model:

• Estimated Regression Model:

tktkttt eXbXbXbbY ...22110

ktkttt XbXbXbbY ˆ...ˆˆˆˆ22110


X

Y

Depiction of Estimation

ttt eXbbY 110

ttt eXbbY 110

X


Estimation Solution

N

t

tt

N

t

t XBBYeQ1

2

10

1

2ˆ

0)(2 10

0

t

tt XBBYB

δQ

0)(2 10

1

t

ttt XBBYXB

δQ

In two dimensions, minimize the objective function with respect to B0

and B1:

Solve the system of equations for B0 and B1

t t

tt XBnBY 10

t t

t

t

ttt XBXBYX 2

10

Reduces to the “normal equations”:Take Derivatives and set equal to zero:


Estimation Solution

In multiple dimensions, parameters are found using matrix mathematics:

BXY

XBXYX TT 11 ))(())(( XXXXBXXYX TTTT

BXXYX TT 1))((

B=

B1

B2

…

Bn

Y=

Y1

Y2

…

Yn

X=

1 X1,1 X1,2

1 X2,1 X2,2

…

1 Xn,1 Xn,2

XT=1 1 … 1

X1,1 X2,1 … Xn,1

X1,2 X2,2 … Xn,2

A solution can be found if the inverse of XTX exists


Advantages of Regression

• Model includes explanatory variables (drivers)

• Parameters provide information about the process

(e.g., price elasticities)

• Forecasts can account for major changes in drivers

• Alternative scenarios can be evaluated

• Caveats:

– Structural changes complicate the forecaster’s work

– You still need to have forecasts of the drivers


Application to TNB Demand

Forecasting

Regression Model


Our Model

• We utilize multiple regression model to forecast our demand which comprises of the following variables:

– Day-of-the-Week Binary Variables

• Mon, TWT, Friday, Saturday

– Monthly Variables

• Jan, Feb, Mac, Apr, May, June, July, Aug, Sept, Oct, Nov

– Holiday Binary Variables

• Major holidays – Raya, CNY

• National holidays – New Year, Christmas, Labour Day

• State holidays – Thaipusam, Sultan’s Birthday & etc.


Our Parameters


Forecast and Performance


Day-ahead Forecast Sample Output

Sample of Day-ahead Daily Forecast Report (23 May 2013)

• Half hourly demand figures from 00:30 to 24:00

• Graph for consistency checks

• Demand summary for maximum demand of the day and daily energy generation requirement

• Three cities weather forecasts used in forecasting assumptions are also summarized in a table


Week-ahead Forecast Sample Output

Sample of Week-Ahead Forecast Report

(Week 22-23 2013)

• Hourly forecast figures for two weeks

• Used for generator outage and transmission outage and production planning


Accuracy and Performance (Daily)

Sample of Day-Ahead Daily Error Analysis (21 May 2013)

• Annual KPI MAPE<2.6%

• Normal MAPE for weekdays (including 0800hrs revision) normally <2.0%

• >95% falls between ±3%


Accuracy and Performance (Monthly)

Sample of Day-Ahead Monthly Error Analysis (April 2013)

• Normal MAPE for weekdays (including 0800hrs revision) normally <1.5%; 1.59% MAPE for Apr ‘13

• >80% falls between ±3%

• Larger error experienced during national and state holidays due to additional uncertainty added to demand characteristics


Accuracy and Performance (Yearly)

Sample of Day-Ahead Yearly Error Analysis (FY2012)

• Comparison between four sets of forecasts produced in FY12

0

1000

2000

3000

4000

5000

6000

7000

8000

-11% -9% -7% -5% -3% -1% 1% 3% 5% 7% 9% 11%

Error Range (%)

Fre

qu

en

cy

Forecast

Rev0800

Rev1600

NDForecast

Error

Summary DA R0800 R1600 NDF

MAPE 1.71% 1.55% 1.51% 3.16%

Energy 1.82% 1.65% 1.61% 3.19%

Daily MD -0.40% -0.31% -0.27% -1.86%

Forecast Rev0800 Rev1600 NDForecast

x<-10% 22 21 21 534

-10%<=x<-5% 343 256 243 1965

-5%<=x<-3% 1466 1163 1052 2906

-3%<=x<-2% 1968 1717 1649 2257

-2%<=x<-1% 2821 2768 2763 2450

-1%<=x<=1% 6631 7518 7688 4341

1%<x<=2% 2155 2202 2246 1335

2%<x<=3% 1221 1095 1101 780

3%<x<=5% 792 733 710 573

5%<x<=10% 144 90 90 268

x>10% 5 5 5 159

Total 17568 17568 17568 17568


THANK YOU

Documents

2. Oc1 _Demand Forecast Electricity